Acceleration of BNCT dose map calculations via convolutional neural networks

Area of Science:

• Medical Physics
• Computational Biology
• Radiotherapy

Background:

• Optimized treatment planning is crucial for Boron Neutron Capture Therapy (BNCT) success.
• Current dose map calculations rely on slow Monte Carlo simulations, limiting treatment plan optimization.
• Accurate dose distribution analysis is essential for effective BNCT outcomes.

Purpose of the Study:

• To develop a machine learning algorithm to accelerate Monte Carlo simulations for BNCT.
• To reduce the computational time required for generating dose maps in BNCT.
• To maintain or improve the accuracy of dose calculations in BNCT treatment planning.

Main Methods:

• A convolutional neural network (CNN) was developed and trained on a dataset of Monte Carlo simulations.
• The CNN was used to accelerate neutron transport simulations for BNCT.
• Performance was evaluated by comparing CNN-generated dose maps against traditional Monte Carlo results.

Main Results:

• The machine learning model achieved high accuracy, with 97% of voxels showing less than 5% error compared to Monte Carlo simulations.
• Inference times were reduced by three orders of magnitude, drastically speeding up the process.
• The CNN model demonstrated its capability to significantly reduce computation time without compromising accuracy.

Conclusions:

• The proposed machine learning approach offers a substantial acceleration of Monte Carlo simulations for BNCT.
• This tool has the potential to enable real-time optimization of BNCT treatment plans.
• The findings pave the way for more efficient and effective radiotherapy planning using AI in medicine.